Skip to content

Models & reasoning effort

How model and reasoning-effort selection work for each connected agent.

Model and reasoning-effort selection are not a separate settings tab — they're two of the four fields on each agent's row inside Settings → Agents, alongside customPath and permissionMode.

Model selection

Where the underlying CLI/SDK exposes a model catalogue, AgentDock lets you pick which model that agent uses for new messages. Model switching support differs by agent:

  • Claude Code and Codex support live model switching within a session, backed by their respective SDKs.
  • Antigravity is PTY/terminal-based rather than a structured SDK transport. Switching its model works by restarting the underlying process and resuming the conversation, rather than switching live in-process.

Reasoning effort

reasoningEffort is a per-agent field that adjusts how much effort/reasoning depth the underlying CLI applies, where that CLI exposes such a control.

For Claude Code, the available levels are low, medium, high, and xhigh / max on models that support the deeper tiers, fetched per-model from the SDK's own catalogue rather than hardcoded — high is the SDK's own confirmed default. Codex passes reasoning effort through per-turn from its own model catalogue the same way. Antigravity's reasoning-effort exposure, where present, goes through its own model-normalization layer — see Antigravity transport.

Note

Exact available levels are model-dependent, not fixed per agent — check the options actually listed for your currently-selected model in Settings → Agents rather than assuming every model on a given agent supports the same tiers.

Related